Model Preparation

Dedicated Altair pre-processor tools for crash integrate a large number of model checkers. These tools also include automatic correction algorithms. The most common checks before running a model are related to:

Mesh quality
Spring definition
Intersections and penetrations in interface definition
Checking the incompatible kinematic conditions:
- No common secondary nodes between rigid bodies
- Only "free" main node
- Spherical inertia for the small ones
Mass and center of gravity
Thickness of the parts
Initial velocity value and direction
Rigid wall normal
Material law: units system consistency
Free nodes
Monitored volume:
- Closed volume
- Positive volume
Connectivity between parts
Mesh sizes between connected parts

Some basic rules to create a valid model are presented.

Mesh Quality

The Mesh Recommendations discussed in Finite Elements should be respected. The shell mesh must be as homogenous as possible. It is not recommended to use different shell formulations for a given physical part. For very fine meshes, it is recommended to use fully-integrated elements or a physical stabilization method (/PROP/SHELL). Triangle shells are avoided. If the mesh includes triangles due to a difficult mesh, it is recommended to reduce its number to 5% per part.

Material Check

Some common material check questions asked are:

Is the material in the right unit system?
- For metallic parts:
  40 GPa < Modulus < 210 GPa
  
  1.8e-6 kg/mm³ < Density < 7.8e-6 kg/mm³
- For plastic parts (PP, PC-ABS, PP GF30%...):
  0.9 GPa < Modulus < 13 GPa
  
  9e-7 kg/mm³ < Density < 1.6e-6 kg/mm³
- For foam:
  1e-8 kg/mm³ < Density < 1e-7 kg/mm³
Is it a proven material from the material database?
Realistic material data needs be input.
Check for these common problems in material.
- Negative slopes in stress-strain curve
- Elastic material being assigned to deformable parts
- Unrealistic yield stress (> 2 GPa)
- Failure is not defined and elements of plastic or metallic parts are stretching unrealistically (plastic strain > 1)
Some of the checks mentioned above can be performed in HyperCrash.
- Unit consistency can be done with the contour check in the quality panel.
- The check “Part is not integrated in rbody for LAW1” finds parts which are deformable and using material LAW1, which can become unstable, if the deformation becomes large.
Figure 1.

Solid and Shell Definition

For some materials it is recommended to use a special option in shell or solid.

For parts using elastic-plastic material law (LAW2, LAW27, LAW36, and so on).
- I_shell=24, I_smstr=4, I_plas=1, I_thick=1, N=5
For solid parts using elastic-plastic material law (LAW2, LAW27, LAW36, and so on).
- For hexahedral elements:
  I_solid=24, I_smstr=4
- For first order tetra elements:
  I_solid=1, I_tetra=0 or 1
- For second order tetra elements:
  I_solid=1
For solid parts using hyperelastic material (LAW42, LAW69, and so on)
- For hexahedral elements:
  I_solid=24, I_smstr=10, I_cpre=1, I_HKT=2
- For tetra elements:
  I_solid=1, I_smstr=10
For foam:
- Material LAW38:
  I_solid=24, I_smstr=10
- Material LAW70:
  I_solid=1 or 17, I_smstr=1 or 11
- For foam modeled with tetra elements, I_tetra should not be set to 1 as foam is highly compressible.
For fabric:
- LAW19 tria elements:
  I_shell=4, I_smstr=1, d_m=0.2, N=1
- LAW58 tria elements:
  I_shell=4, I_smstr=4, d_m=0.2, N=1

This can also be checked with HyperCrash Model Checker.

Spring Definition

Non-physical or bad definition of springs is a common problem in crash models. In fact, the properties of springs must be consistent in mass, inertia, stiffness and length. Actually, a spring must have a physical behavior:

With regard to mass distribution over a one-dimensional bar, the inertia will have upper and lower limits: (1)
$\frac{m l^{2}}{12} \leq I \leq \frac{m l^{2}}{4}$

The lower limit is the inertia of a uniform mass distributed bar. The upper limit gives the inertia of a bar with two extremity masses m/2. As the springs are also used to model bolts and spot welds, you can go beyond these limitations and write:(2)
$0.01 m l^{2} \leq I \leq 100 m l^{2}$
The cross-section of a spring can be computed either by $S = \frac{K l}{E}$ or $S = \frac{m}{ρ l}$ . If the ratio between the two computed values is greater than 100, the inconsistency may result some trouble.
In the spring property, for a negative strain the force must be negative and for a positive strain, it must be positive. Otherwise, the spring generates energy during computation.
If the tangent stiffness is negative for a nonlinear elastic spring, there is a risk of instability especially if vibrating in the negative slope zone (energy error may become positive and increase).
If the maximum slope of the curve (so the maximum stiffness) is greater than the initial stiffness, unloading in the zone of maximum slope will be false (see Modeling Tools).
Very high stiffness in spring elements can cause low timestep and instability.
For example, the following check in HyperCrash allows to identify the spring element that have unusually high stiffness in tension (1000 kN/mm) and rotation (10000 kN.mm/rad).
Note: Property that is referenced in parts of the dummy / barrier models should be ignored.

Figure 4.

Beam Consistency

In addition to the beam assumption length/depth > 10, the following recommendations define a consistent beam:(3)

L > \sqrt{A}

$\begin{array}{l} 0.01 A^{2} < I_{y} < 100 A^{2} \\ 0.01 A^{2} < I_{z} < 100 A^{2} \\ 0.1 (I_{y} + I_{z}) < I_{x} < 10 (I_{y} + I_{z}) \end{array}$

Intersections and Penetrations in Interfaces

Initial mesh intersections create unrealistic connections. Moreover, this can cause locking situations leading to computation failure.

Initial penetrations are due to the interface gap definition. They lead to unrealistic and uncontrolled internal forces which may cause local plastic strains in the beginning of the computation. The structure is then locally less stiff.

Altair pre-processor tools can be used to detect and remove intersections and initial penetrations. A model containing many initial penetrations cannot be considered as valid. After corrections, if a few number of initial penetrations remain, the flag Inacti =1 can be activated in /INTER/TYPE7. In this case, the interface stiffness for the nodes initially penetrated is deactivated.

If only small initial penetrations remain (less than 5% of the gap), the variable gap in time can be used by setting the flag Inacti=5; which is better than Inacti=1.

Best practice (recommended):
- For interface TYPE7, 11 and 19:
  - I_stf=4, St_min= 1, I_del = 1 or 2, Inacti=6, I_form=2, Gap_min ≥ 0.49
  - There is no intersection in the areas where the simulation fails
  - Penetrations are not too deep (check residual distance in HyperCrash/HyperMesh). It should not be below 0.1mm.
- For interface TYPE2:
  - Use Spot_flag= 28 if main is shell or secondary node has rotational DOF and main does not (solid)
  - Use Spot_flag= 27 if both secondary and main do not have rotational DOF (solid)
  - I_del should be set to 1 for all cases

Rigid Body Definition

The following recommendations should be pointed out:

Never use a node of the mesh as a main node for a rigid body
A rigid body made of two secondary nodes requires using the spherical inertia flag (I_spher =1 in /RBODY)
Be aware that the main node may move when initializing the rigid body in Radioss Starter

Most rigid body definition problems can be detected and corrected using Altair pre-processor tools.

Incompatible Kinematic Conditions

Incompatible conditions may results in an increase of the total energy and non-reproducibility of results in parallel computation even if /PARITH/ON is used. Radioss Starter detects the potential incompatible conditions that must be controlled by you.

Parallel Computation

The option /PARITH/ON must be used when using multi-processors. If it is not used, the problems will be difficult to reproduce and may appear and disappear.

Comments

Using Inacti =1 makes the model softer.
Never put free nodes in the interfaces. The selection of surfaces in interface definition must be done with shells and solids elements (or parts) and not nodes.